Abstract
A systematic inter-hemispheres edge-based structural similarity analysis between a healthy brain and various alterations in brain anatomy induced by some neurodegenerative diseases in brain MR images is reported. The edge-map of the brain shows potential as an early predictor of disease-onset and prognosis. The present study has determined the best edge-detection operators that are suitable for edge-based structural inter-hemispheres similarity analysis and aims to illustrate how the edge-map analysis could be a discriminative tool in the case of the studied diseases. The results provided by consecrated edge detectors and newly proposed structural filters have been carefully corroborated with the edge-based structural similarity (ESSIM) and feature-similarity (FSIM) indexes that allow an inter-hemispheres similarity comparison. The multicolliniarity analysis has been used to check the ability of the ESSIM and of the FSIM to discriminate between the neurodegenerative diseases and the normal aging process of the brain. Finally, beta-index allows for edge-preservation estimation. Although this kind of analysis has been the subject of many studies to date, the objective inter-hemispheres similarity measures are considered with the aim of optimization of edge-detection operators, and thus is the applicability of the selected filters increased. The assessment of the stability of the similarity with the help of multicollinearity represents the novelty of our work. A number of 115 2D MR images of the normal brain aging and three degenerative diseases have been selected. Experimental evaluation has clearly demonstrated that the best filters are Sobel, Prewitt, LoG and Scharr as they better preserve the image features and allow the analyzed classes to remain independent.